Choose one main AI system and a second for a specific task — do not spread yourself thin
22 episodes in 12 months · 25 total · first in August 2025
I am not tied to any one company and I move to wherever is objectively better. I split the work by role: analysis and requirements go to ChatGPT, development to Claude and Codex.
But ten systems at once is spreading yourself thin: even working in two is hard already. Choose one main model and keep a second for a specific task, such as development.
Happening now. Right now I have a division of roles: about 80% of development is in Claude, the rest is ChatGPT and Codex (No. 154, August 2026).
Where to start
Choose a main model and one more for a specific task — and do not spread yourself over ten subscriptions.
Do not tie your processes to one model for good: the leader changes, so move to wherever is objectively better.
You cannot handle more than two systems: even working in two is already incredibly hard.
“And actually, even working in two systems is already becoming incredibly hard, practically unrealistic.”
When I said it
Quotes
“I said that if for me the main tool right now is ChatGPT, mostly I use Gemini, I use Grok, but mostly ChatGPT”
“Right now, roughly, well, when developing serious systems I already use three systems like that. I mean, I work simultaneously in ChatGPT 5 Pro, 5.5 Pro, constantly, like, an-analysing systems, repositories, uh, prompts, evaluations.”
“But in terms of programming, obviously, for a month now 80% of my work has been happening in Claude, not in ChatGPT. I continue to use Codex a lot for different tasks.”
“If two and a half years ago you could still, like, run tests in ChatGPT, and in Claude, and in DeepSeek, even, probably, in Gemini, then as of today you have to focus yourself very clearly. If you can use one model, but using one, say, ChatGPT or Claude, whichever you like, yes.”